Pay per use · No subscriptions

Honest, per-token pricing.

Top up once. Spend at per-model token rates. Same prices for everyone, from your first request to your millionth. No monthly fee. No lock-in. Credits never expire.

Free tier
$1 credit on signup · all models · 20 requests / min

No card required. Try every model end-to-end before you spend a cent. Top up to unlock 600 RPM and 1M tokens/min.

Start free

Per-token pricing

All prices in USD per million tokens. Cached prompts and embeddings are billed separately at lower rates.

8 chat models · 1 embedding
ModelFamilyInput / 1MOutput / 1MContext (tokens)
translategemma-4bgemma$0.05$0.2131,072
qwen3qwen$0.1$0.332,768
deepseek-ocrFeatureddeepseek$0.2$0.88,192
llama3.1Featuredllama$0.2$0.4128,000
gemma3Featuredgemma$0.3$0.68,192
deepseek-r1deepseek$0.3$0.932,768
gpt-oss-120bFeaturedgpt-oss$0.4$1.2131,072
qwen3-coderFeatureddeepseek$0.6$1.832,768
nomic-embedEmbeddingembedding$0.058,192

Prices are billed per token of completed request, rounded to the nearest millionth of a dollar. We do not charge for failed requests.

What $1 actually buys
gemma4
10M input or 3.3M output tokens
About 7,500 short chat completions
qwen3-32b
2M input or 670K output tokens
About 1,500 short chat completions
qwen3-coder
1.6M input or 550K output tokens
About 1,200 coding completions

Top-up packs

One-time purchases. Credits never expire. Pay with card or Bitcoin.

Indie
$10credit

Kick the tires

Sign up
Builder
$25credit

Ship a side project

Sign up
Most chosen
Team
$50credit

Run agents for weeks

Sign up
Scale
$100credit

Production workloads

Sign up
Enterprise
$500credit

Volume discount

Sign up

Credits never expire

Top up once, use over months. We don't burn your balance for inactivity.

Hard spend caps

Per-key daily and monthly caps stop runaway agents before they drain your balance.

Same price for everyone

No volume tiers, no enterprise tax. Your first request costs the same as your millionth.

Start with $1 of free credit.

No card. No subscription. Just an API key and 8 self-hosted open models.